On 11/19/2013 12:59 PM, Josh Berkus wrote:
> On 11/19/2013 08:14 AM, Robert Haas wrote:
>> On Thu, Nov 14, 2013 at 2:54 PM, Hannu Krosing <hannu@2ndquadrant.com> wrote:
>>> I am sure you could also devise an json encoding scheme
>>> where white space is significant ;)
>> I don't even have to think hard. If you want your JSON to be
>> human-readable, it's entirely possible that you want it stored using
>> the same whitespace that was present on input. There is a valid use
>> case for normalizing whitespace, too, of course.
> Given that JSON is a data interchange format, I suspect that there are
> an extremely large combination of factors which would result in an
> unimplementably large number of parser settings. For example, I
> personally would have use for a type which allowed the storage of JSON
> *fragments*. Therefore I am interested only in supporting two:
>
> a) the legacy behavior from 9.2 and 9.3 so we don't destroy people's
> apps, and
>
> b) the optimal behavior for Hstore2/JSONB.
>
> (a) is defined as:
> * complete documents only (no fragments)
> * whitespace not significant
> * no reordering of keys
> * duplicate keys allowed
>
> (b) is defined as:
> * complete documents only (no fragments)
> * whitespace not significant
> * reordering of keys
> * duplicate keys prohibited
>
> If people want other manglings of JSON, they can use TEXT fields and
> custom parsers written in PL/v8, the same way I do.
>
Fragments are currently allowed in a):
andrew=# select '"a"'::json; json ------ "a"
Given that, I'm not sure we shouldn't permit them in b) either. I think
I lost that argument back in the 9.2 dev cycle. I really don't want to
get to a situation where foo::json::jsonb can produce an error.
cheers
andrew